With all the market interest in artificial intelligence, it’s no surprise that many are asking about the best way to learn more about it. What should I read? What should I watch? There’s so much material out there. But, before one can properly answer those types of questions, it’s useful to take a step back and consider what “doing AI” even means because it turns out that AI can mean a lot of different things depending upon what you’re trying to accomplish.
In this talk, Gordon Haff will provide you with both a high-level roadmap and specific pointers for adding AI smarts to your toolbox. He’ll distinguish between research AI and applied AI, discuss how AI intersects with data science more broadly, and look at some of related research and practice areas that will help you understand AI beyond just machine learning. Armed with this knowledge, you will be better prepared to chart out a program for learning AI that targets your specific needs and objectives rather than wasting time on topics that are not interesting or relevant to you.
Nell’iperspazio con Rocket: il Framework Web di Rust!
The good the bad and the ugly: Getting started doing AI
1. 1
The good the bad and
the ugly: Getting
started doing AI
Gordon Haff
Technology Evangelist
@ghaff
2. 2
Who am I?
● Evangelist for emerging
technologies and practices at Red
Hat
● Author of How Open Source Ate
Software, etc.
● Former IT industry analyst
● Former big system guy
● Website: http://www.bitmasons.com
5. An AI Map
Research Applied
Machine Learning
Deep
Learning
Brain science &
Cognitive psychology
Linguistics &
NLP
Human/machine
interactions
Supervised learning
Unsupervised learning
Reinforcement
learning
Domain
expertise
Robotics
Data
anonymization
Data science &
statistics
AI
6. 6
Research AI
● Math heavy (linear algebra,
calculus, optimizations,
probability)
● Essentially university
curriculum
● Can touch many adjacent areas
● Not necessarily primarily
programming/working with
data
7. 7
Research AI resources
● Many MOOCs/university courses/text books
○ AI, Machine Learning, Deep Learning
○ Foundational courses such as linear algebra and calculus
○ Adjacent fields such as cognitive psychology and linguistics
● Other open educational resources (e.g. MIT
OpenCourseWare)
● Research papers
8. 8
Applied AI
● Applications that solve today’s
problems
● Background in relevant statistics
and algorithms
● Programming
● Data science stuff (data
cleansing, presentation, etc.)
● Primarily makes use of ML/DL
9. 9
Machine Learning
Machine learning is a method
of data analysis that
automates analytical model
building. It is based on the
idea that systems can learn
from data, identify patterns,
and make decisions with
minimal human intervention. https://www.geeksforgeeks.org
10. 10
Deep Learning
● Sub-set of machine learning
that uses multi-layer neural
networks
● Has been the primary
approach that has led to so
many recent “AI” advances
● Beneficiary of increased
computation/data, including
accelerators such as GPUs
14. 14
But reinforcement learning limits
● Learn from mistakes
● Physical world versus models
● Exploration versus exploitation
● Real-world environments change
● States can be poorly defined
18. 18
Amazing stuff since ~2010
● Voice recognition: Siri, Alexa, Cortana, Google
● IBM Watson wins Jeopardy
● Computer vision classification can beat humans
● Autonomous driving research
● Ubiquitous bots
● Lots of unsexy predictive analytics, trading,
optimization, and analysis
19. 19
Heathcare Example: ChRIS
● Real-time Web-based MRI Data Collection,
Analysis, and Sharing
● Cloud-based platform developed as part of a
collaborative effort between Boston Children’s
Hospital, Red Hat, Boston University, and the
Open Cloud (MOC)
● Began as a way to facilitate the organization, 3D
visualization, and collaboration around medical
imaging amongst researchers
20. 20
Supervised learning challenges
NO PHYSICAL WORLD CONTEXT
● Lack of real world context
● Interpretability
● Dependent on large
training sets
● Sensitive to small changes
21. 21
The basics
● Programming & programming environment
○ Programming for Everyone, UMich
(Python)https://online.umich.edu/courses/programming-for-everybody-getting-started-with-python/
○ Introduction to Computer Science and Programming using
Python, MIT
https://www.edx.org/course/introduction-to-computer-science-and-programming-using-python-2 (Text is Introduction
to Computation and Programming using Python by John Guttag)
○ Anaconda distribution (Python/R/TensorFlow/data science
libraries/Jupyter notebooks)
○ SQL https://www.khanacademy.org/computing/computer-programming/sql
○ Sabermetrics 101: Introduction to Baseball Analytics on edX
is a fun and gentle introduction to data analysis
22. 22
Data Science: Working with data
● Python for Data Analysis, O’Reilly
● Kaggle
● MicroMasters in Statistics and
Data Science, edX (MIT)
https://www.edx.org/micromasters/mitx-statistics-and-data-science
● CS109 Data Science, Harvard
http://cs109.github.io/2015/pages/videos.html
http://blog.operasolutions.com/bid/384900/what-is-data-scienc
23. 23
Deep Learning
● Deep Learning by Ian Goodfellow et al.
https://www.deeplearningbook.org/
● Good list of deep learning resources
https://blog.floydhub.com/
● More practically-grounded courses
(MOOC/YouTube/fast.ai), e.g. MIT 6.S094: Deep
Learning for Self-Driving Cars
24. 24
“Democratized” AI
● Cloud AI/ML services like Google Cloud AutoML
(and cloud generally)
● “Cookbooks,” e.g. O’Reilly Deep Learning Cookbook:
Practical Recipes to Get Started Quickly
● Python libraries, Jupyter notebooks
25. 25
Keep your eye on
● Value/utility of data vs. privacy: MPC,
homomorphic encryption, etc.
● Ownership of data
● Voice interfaces
● Explainability and bias
● Beyond current deep learning
● Multidisciplinary work